| With the development of Internet based on TCP/IP, the scale, users and traffics of Internet have experienced an explosive growth since 1990's. The network congestion has become more serious and complex due to the ever-increasing network application types. Congestion often results in decline of quality of service(QoS) in terms of transmission delay and throughput, which affects the utilization of network resource like bandwidth and buffers. So the schemes of congestion control are very important for the robustness and stability of the networks, and the congestion control is always a hot spot in the field of network research.Currently,there are two kinds of the schemes of TCP/IP network's congestion control, the congestion control strategies based on sources and the strategies based on link algorithm. The two control schemes are impacted on each other. Congestion control schemes of TCP/IP Networks thoroughly researches in two different ways of sources and routers. The main research work about this dissertation is as follows:1. To solve the problems of the UDP and TCP application of sharing network resources fairly, we mainly adopt the following two ways to put forward the strategy of congestion control: The first is the routers implemented the mechanism of active queue management. The routers monitor the queue for each connection receiving data. In order to share the bandwidth fairly to all connections, it punishes the connections occupying more the share of the link than the ones having no congestion control and continuing to send data. The other way is that add congestion control algorithm to UDP protocol based on the model of TCP sending rate, and emphasis TCP-friendly control on rate adjustment of the sources. So it makes the multimedia applications share with TCP in the same network, get a fairly bandwidth, and improve the data transmission quality of network.We provide the following control strategies according to the above two method. Firstly, a NTFFP algorithm is proposed to punish non-friendly TCP flows based on packet drop history in the routers. The algorithm will record the bandwidth of those active connections that have packets currently in the buffer, and utilize the character of the ratio of packet arrival rate and total arrival rate from a particular connection is roughly proportional to that of packet drop numbers and total packet drop numbers in the buffer. It punishes those flows with an excessive share of bandwidth, so that it reaches the effect of sharing the bandwidth in the different data flows. The simulations validate that NTFFP can raise the punishment strength to non-friendly TCP flows, increase the fairness of TCP, as well as keep high link utilization.Secondly, the end hosts and the routers in network as a whole are regarded as a feedback control system, the UDP multimedia applications have the congestion control mechanism, and dynamically adjust the sending rate according to the feedback status of the network. The ECN-based and Enhanced-fairness TCP-friendly congestion control is proposed for TCP friendly congestion control. It resolves the problem of UDP flows having no congestion control. The routers early notify to the sender using ECN as congestion indication, and can response to congestion fleetly. This mechanism adopt ERED algorithm to mark the arriving packets with a probability, and improve the fairness problem of the flows have different packet sizes in the router. The applications based on UDP protocol can friendly work in peace with TCP flows. The simulations validate that EETCC flows can share the bandwidth of bottle-neck links with TCP flows, and adjust smoothly the sending rate. So the algorithm can reduce the loss rate, increase the link utilization, and present the good TCP-friendly.2. We focus on the Active Queue Management (AQM) strategy on the link. At present, TCP / AQM linear model will lead to a substantial queue oscillation, and increase the probability of empty queue. The phenomenons increase the end-to-end delay jitter and reduce the link utilization. As a result, this article designs the new AQM control strategies to satisfy the requirement of stability, robustness and rapid response from the perspective of control theory.Firstly, by analyzing the TCP / AQM feedback control model, we can see that the time-delay links will lead to instability in the control system. The paper is proposed that a new AQM algorithm called SPI which combined Smith predictor and Dahlin algorithm. When designing rules, using the dynamic nature of the predictor object, we adopt a predictor model to compensate the time delay, so that move the time delay term to the outside of feedback control loop. SPI can eliminate the influence of time delay on the system performance and improve the stability and the rapid response of the system. We give the condition of the parameters established in the stable system. The result of simulation shows that the presented SPI controller can enhance the performance more remarkablier than PI controller, keep the queue length at the desired value, and have better stability.Secondly, considering the complexity of network traffic modeling, we explore the self-adaptive to realize AQM. So the neural networks will be introduced to control congestion. By the single neuron adaptive controller adjusting the weight on-line, the algorithm obtains self-learning and self-adaptive ability. The weighted factors are regulated continuously according to the system errors. Therefore the algorithm speeds the convergence of the queue, has the ability of adapting to the network environment change, and realizes the on-line control of the controlled object. It adopts the method of minimizing the objective function to improve the stability and robustness in the dynamic network. We give the change range of the adjusted parameters in the paper. The result of simulation shows that the presented ANPID controller can achieve the performance indicators of active queue management algorithm, as well as stabilize queue length to the desire value quickly.3. Congestion control is an essential problem in wireless sensor networks. When the wireless sensor network occured congestion, there are two problems to be solved: First, to reduce the data sending rate of the nodes in the congestion regional, in order to ease congestion. It will reduce the data flows to the network capabilities, and prevent the packet loss. The second is the fairness and reliability of network control, the aim is to ensure that all data source have the same (or the weight) right to access the network bandwidth, and to ensure the reliability of data transmission.Traditional wireless sensor network congestion control algorithms are usually at the expense of fairness and reliability. In order to achieve congestion control, these algorithms decrease the flow by reducing the data sending rate. To solve these problems, the optimization theory will be introduced to wireless sensor network congestion control. An optimization rate-based congestion control algorithm is proposed, which builds a distributed network framework, and adopts the utility function to express the function of the transmission rate of the cluster nodes. The algorithm makes the sum of the utility function of the cluster nodes maximize as the optimization objective function, plans reasonably the flow rate distribution of wireless sensor nodes from the scope of network, and ensures the reliability of data transmission. The simulations validate that ORCC can regulate continuously the data transmission rate of the sources based on the congestion estimation. The algorithm can reduce the average delay, achieve the balanced state of system, and realize the reliability and fairness of network transmission between the nodes.Summarily, this paper accomplishes some foundational theory research work for the congestion control stragety of the TCP/IP network and the wireless sensor network. The main respect of this dissertation is innovation and feasibility of some algorithms realizing the system performance, and some algorithms are verified in the simulation platform .
|
PDF Full Text Request